[U-Boot] [PATCH 00/11] 85xx/86xx dma updates

[Ira Snyder]

On Thu, May 21, 2009 at 03:49:29PM -0500, Peter Tyser wrote:
> On Thu, 2009-05-21 at 12:37 -0700, Ira Snyder wrote:
> > On Thu, May 21, 2009 at 12:09:58PM -0500, Peter Tyser wrote:
> > > This patch series attempts to clean up the DMA implementation for the
> > > 85xx and 86xx architectures.  The changes include:
> > > - consolidate 85xx and 86xx structures and code
> > > - add defines for bitfields
> > > - use proper IO accessors
> > > - add support for arbitrarily large transfer sizes
> > > - rename dma_xfer() to dmacpy() and make dmacpy's prototype similar
> > >   to memcpy()
> > > 
> > > The patches are based on the mainline "next" branch.
> > > 
> > > I've tested the code on MPC8572 and MPC8640-based boards.
> > > 
> > > I'm not initimately familar with the 83xx platform, but at a glance
> > > it looked like the fsl_dma driver could be extended to support it
> > > with some ifdeffery.
> > > 
> > 
> > I've been doing some Linux work with the 83xx DMA controller. The only
> > real differences between the 83xx/85xx controller are the register
> > endianness and snoop bits. The 83xx controller's registers are all
> > little-endian, while the 85xx is all big-endian.
> > 
> > Also, there are some snoop bits that need to be enabled on 83xx as well,
> > in the control register as well as in each descriptor if you're running
> > in chaining mode.
> > 
> > That's everything that I've noticed that is different. I don't have an
> > 85xx/86xx to test anything with, but I'm happy to run some tests on my
> > mpc8349emds if you want to try adding support for 83xx.
> 
> Thanks!
> 
> I believe some of the register locations are slightly different too, for
> example the source address on the 85xx is at offset 0x114 while its at
> 0x110 on the 83xx.
> 
> I don't think any 83xx boards currently use the 83xx DMA implementation
> in cpu/mpc83xx/cpu.c.  Before spending any time on the 83xx, is there
> any good reason to support the 83xx in U-Boot?  There would be no users
> of the updated implementation as is.
> 

I think there is an option for the mpc8349emds to use the DMA controller
to initialize the RAM for ECC mode. I tried using it, and it was much
slower than using the CPU.

I'll probably end up writing a virtual network driver for U-Boot
eventually. In the Linux version, I've used the DMA controller to handle
transferring data over PCI. It is much faster than the CPU, but I'll
only be transferring a few megabytes from within U-Boot anyway.

In short, having the driver ported to 83xx would make some of my future
development easier, but there aren't any serious in-tree users at the
moment (to the best of my knowledge).

Ira